Solar sites have hills. Most wind models pretend they don't.

Ask how the wind loads were derived on your last utility-scale project and the answer is usually some version of the same thing: code values, a terrain category, a roughness assumption. Flat ground, adjusted.

Real sites are not flat ground adjusted. A hill compresses the flow over its windward slope, accelerates hard at the crest, then drops into a deep velocity deficit on the lee side. Two rows of trackers 200 metres apart can sit in completely different load regimes. Averaging that away doesn’t remove the risk. It just moves it off the spreadsheet.

Wind load is one of the largest single drivers of steel cost on a solar project. Get it wrong optimistically and you carry structural risk into a 30-year asset. Get it wrong cautiously and you overbuild every pile and torque tube on the site, which prices you out of the tender before the design is even reviewed.

The obvious fix, a wind tunnel analysis, costs weeks the schedule doesn’t have. Which is why the question worth answering is whether simulation can actually reproduce what terrain does to wind, and how anyone would know.

The hardest terrain in wind engineering

NablaFlow has published a full validation report for ArchiWind’s standard CFD workflow, authored by Knut Erik T. Giljarhus. The terrain half of the study uses the two most demanding public benchmarks available.

Askervein Hill, Scotland. A 116-metre isolated hill instrumented with more than 50 measurement towers during the 1982–83 field campaigns. It remains one of the most thoroughly documented full-scale experiments of atmospheric flow over terrain anywhere in the world.

Bolund, Denmark. A 12-metre hill with a near-vertical escarpment on its western face. Small, but brutal: it produces sharp crest acceleration plus flow separation and recirculation in the lee. Considerably more demanding than smooth terrain.

Both were run through the same automated ArchiWind workflow a customer uses. No hand-tuning, no case-specific fixes.

The results

At Askervein, ArchiWind reproduced the full sequence along both measurement lines: the initial slowdown at the foot of the hill, the acceleration up the windward slope, the crest maximum, and the strong lee-side deficit.

At Bolund, the predicted speed-up fell within the experimental uncertainty range at every measurement point, including in the separation region behind the escarpment the part of the flow that defeats most steady-state models.

The report is equally clear about the limits. At the Askervein crest, the simulated peak is slightly lower and smoother than the measured one. That characteristic is documented rather than buried.

What this changes when you’re scoping a site

Terrain-aware wind modelling stops being a research exercise and becomes a procurement input. You identify the amplification zones before the structural design is frozen, rather than discovering them at commissioning. Nextpower already runs this workflow across it’s pipeline, and the methodology in this report is what sits behind it.

The report also compares ArchiWind’s two mesh settings. Draft runs on a coarser mesh for speed; Detailed refines around the terrain and buildings. On this benchmark, both produced effectively the same global accuracy. In practice that means you can screen a portfolio of candidate sites at Draft speed, then commit compute to Detailed only on the sites that make the shortlist. The report is clear about the trade-off: a standard Draft study covers eight wind directions rather than sixteen, so it’s a screening tool, not a substitute for a final assessment.

Click here to read the full validation report